What this category of data looks like
The data for this category comes from slab/billet batches, heating furnaces, mill automation control systems, online thickness and width measuring equipment, surface inspection systems, and MES systems. Daily documents include rolling plans, heating schedules, rolling specifications, plate shape and thickness records, surface quality reports, and quality certificates. Fields include parameters with clear units such as batch number, thickness (mm), width (mm), heating temperature (℃), and rolling speed. Data is updated in real time during each coil or individual steel billet production cycle, with each single record corresponding to the full-process process parameters of a single batch or coil of steel.
Constraints on Knowledge Base Retrieval and Recall
The real-time coil-level update requirement for this category means the knowledge base retrieval must support incremental small-batch indexing, to avoid resource consumption from full re-indexing. Process parameter fields with clear units require unit information to be matched during retrieval, to prevent result deviations caused by fuzzy matching without units. Multi-source heterogeneous data sources require retrieval to support mixed queries of structured parameters and text descriptions. Equipment inspection and repair questions mostly involve specific parameters of a single piece of equipment or a single coil of steel, so recall results must accurately match the question scenario and avoid interference from irrelevant mass production data.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.9 | Steel rolling process parameters have clear thresholds; this interval filters irrelevant results with low matching degrees |
limit | Top 8–12 results | Equipment inspection questions usually involve parameters of a single piece of equipment or a single coil of steel; this range balances recall volume and context load |
searchMode | mixedRecall | Documents in this category include structured parameters and text process descriptions; mixed retrieval balances accurate matching and semantic understanding |
usingReRank | Enabled | Rolling data has strong process relevance; re-ranking improves the ranking priority of highly relevant results and reduces misjudgment probability |
datasetSearchUsingExtensionQuery | Disabled | Equipment inspection questions mostly target specific process parameters; extended queries may introduce irrelevant mass rolling plan data |
collectionIds | Specify dedicated knowledge base ID for rolling and finishing | Isolate documents from other categories to avoid irrelevant data such as steelmaking and forging from mixing into retrieval results |
This configuration was tested and verified on FastGPT v4.17.0.
Three Common Misconfigurations
- Symptom: The dataset remains in the "Indexing" state for a long time after import, with no progress updates. Cause: No incremental indexing trigger rule is configured, or the single-batch imported data volume exceeds the platform's single-processing limit, causing indexing tasks to block.
- Symptom: The similarity value returned by retrieval is abnormally high, such as 10000+. Cause: The
similaritythreshold is not set correctly, or parameter units are not matched during retrieval, causing the embedding model to perform incorrect matching on fuzzy text without units. - Symptom: The equipment inspection workflow interrupts halfway and cannot complete the AI conversation step. Cause: No dedicated
collectionIdsis specified, and a large amount of irrelevant mass production data mixes into retrieval results, causing context overflow or request timeout.
How to Verify Proper Configuration
- Upload the plate shape and thickness record document for a single coil of steel, initiate an inspection question that includes specific parameters, and verify that returned results accurately match the corresponding rolling specifications.
- Check the dataset indexing status, confirm that small-volume documents (single-coil data) can complete indexing within 10 minutes with no long-term blocking.
- Initiate a test question with clear units, such as "What are the rolling reduction parameters for 15mm thick steel plates", and verify that the
similarityvalue of the returned results falls within the0.75–0.9interval. - After enabling
usingReRank, verify that the process logic of the top 3 retrieval results matches the question, with no obvious irrelevant content.
The values provided here are common starting points and should be evaluated against the reader's own samples.